---
id: "2026-10-14-baker-lab-de-novo-designed-proteases-nature"
url: "https://postcutoff.com/e/2026-10-14-baker-lab-de-novo-designed-proteases-nature/"
as_of: "2026-10-14T22:23:00+02:00"
date: "2026-10-14"
date_precision: day
category: science
importance: 3
confidence: high
status: [Result confirmed]
verification: Peer-reviewed
sources: 4
editor: human editor
human_review: null
version: null
---

As of: 2026-10-14 22:23 CEST.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/e/2026-10-14-baker-lab-de-novo-designed-proteases-nature/

# AI-designed proteases (Baker lab, Nature)

Full title: Two Nature papers from the Baker lab: proteases designed from scratch with RFdiffusion cut chosen protein sequences

On Oct 14, 2026 Nature published two papers from David Baker's lab describing enzymes that break peptide bonds, designed de novo with RFdiffusion-family tools: cysteine proteases (13 of 69 designs active, 19%, rate enhancement up to 3x10^7, folds unlike any natural protease) and zinc metalloproteases (36% of 135 designs active; the best, after four substitutions, above 10^10-fold rate enhancement) that cleave chosen sequences, including in disease-associated human proteins. Peptide bonds are stable for hundreds of years in water, so this is a much harder design target than the low-barrier reactions earlier de novo enzymes handled.

## Key facts
- Paper 1: 'De novo design of cysteine proteases' (Nature, online Oct 14, 2026; first author Hojae Choi, senior authors Samuel J. Pellock and David Baker). Success rate 13/69 = 19%; kcat/kuncat up to 3x10^7; designed folds unlike natural proteases (TM score < 0.50); six crystal structures within 1.2 A Ca RMSD of the design models
- Paper 2: 'Computational design of metalloproteases' (Nature, online Oct 14, 2026; first author Anqi Chen, senior authors Donald Hilvert and David Baker). Of 135 designs, 36% cleave their intended peptide substrate at the targeted bond; the best accelerates hydrolysis more than 10^8-fold, and four substitutions take it above 10^10-fold
- Targets: designed metalloproteases selectively cleave the human disease-associated proteins TDP-43, amyloid-beta and serum amyloid A, with rate enhancements up to 10^9-fold; the paper also reports caged cytokines and receptor antagonists activated by the designed proteases (abstract)
- Pipeline (Ars Technica): RFdiffusion2 generates a backbone constrained to hold the catalytic residues and a target sequence; ProteinMPNN fills in the amino acids; AlphaFold3 predicts structures. Starting points were a crayfish digestive zinc protease and a yeast cysteine protease
- Ars Technica's reading of the zinc work: designs that did not place a catalytic water often failed (14 of 89 weakly active), while adding the water as a design constraint raised this to 35 of 38
- Limitation stated by the authors (via Ars Technica): the tools can be told to satisfy structural constraints, not to achieve a specific catalytic activity, so designs are generally far less active than evolved enzymes; protein delivery and immune response are open obstacles to any therapeutic use

## What happened

Both papers start from a natural protease's catalytic core and replace everything else with a computationally designed
protein. RFdiffusion2 builds a backbone that holds the catalytic residues in place and binds the intended substrate
sequence, ProteinMPNN chooses the amino acids, and AlphaFold3 checks the predicted structures. Crystal structures match the
design models closely, and the designed folds do not resemble natural proteases.

## Why it matters

Earlier de novo enzyme design worked mainly on easy reactions. Cutting the peptide backbone at a chosen sequence is a long-
standing target because it would allow proteins to be made and then selectively destroyed or activated. The disease-protein
demonstrations are proof of principle only; delivery and immunogenicity remain unsolved.

## Your AI and this story

- GPT-6 Astra (training cutoff April 2026): 167 days after its cutoff
- Claude Opus 5.5 (training cutoff June 2026): 106 days after its cutoff
- Gemini 4 Argon (training cutoff March 2026): 197 days after its cutoff
- Grok 4.7 (training cutoff May 2026): 136 days after its cutoff

## Sources

1. [Nature: De novo design of cysteine proteases (Oct 14, 2026)](https://www.nature.com/articles/s41586-026-11143-w) (nature.com, paper)
2. [Nature: Computational design of metalloproteases (Oct 14, 2026)](https://www.nature.com/articles/s41586-026-11167-2) (nature.com, paper)
3. [Institute for Protein Design: RFdiffusion2, efficient enzymes from prompts](https://www.ipd.uw.edu/2025/12/rfdiffusion2-efficient-enzymes-from-prompts) (ipd.uw.edu, official)
4. [Ars Technica: Researchers use AI to design proteins that cut up other proteins (John Timmer, Oct 14)](https://arstechnica.com/science/2026/10/researchers-use-ai-to-design-proteins-that-cut-up-other-proteins/) (arstechnica.com, press)

## Changes

- 2026-10-14 (filed): Created from both Nature abstracts and Ars Technica's write-up

## Related

- 2026-10-05: [Google DeepMind's AlphaProtein Novo designs new-to-nature enzymes from scratch (piperidine synthesis, DEHP degradation), with Frances Arnold's lab](https://postcutoff.com/e/2026-10-05-deepmind-alphaprotein-novo-enzyme-design/index.md)
- 2025-11-05: [Baker lab designs antibodies from scratch with atomic accuracy using RFdiffusion](https://postcutoff.com/e/2025-11-05-ai-designed-antibodies-rfdiffusion/index.md)
- 2023-07-11: [RFdiffusion: diffusion models design new proteins that work in the lab](https://postcutoff.com/e/2023-07-11-rfdiffusion-protein-design/index.md)